Gary Marcus

Rank 13 of 47
|
Score 84
6/27/2025, 9:56:04 PM
X
In reply to:
Vincent Paul
@_vincentpaul_
·
229d
@GaryMarcus @eshear Perhaps the better analogy is that a knife can occasionally slip and cut you, if you’re not using it correctly?
Gary Marcus
@GaryMarcus
·
230d
@_vincentpaul_ @eshear it’s like discovering that your knife can suddenly jump off the table.

if you read the paper, the outcomes are very very far from the training. even the authors didn’t expect the specific result.
Vincent Paul
@_vincentpaul_
·
230d
@GaryMarcus @eshear Curious about your take. The study suggests narrow fine-tuning on malicious tasks can broadly misalign LLMs. Interesting, but a "No duh" moment for me? Like blaming a knife because it can cut. Powerful tools inherently carry risks; thoughtful use and design remain key.
Gary Marcus
@GaryMarcus
·
230d
Vile.

ChatGPT should be pulled from the market til they fix this.

The statement is part of a broader discussion on the unexpected outcomes of AI models, specifically regarding safety and alignment issues. The conversation critiques the reliability of AI systems and their potential risks, using analogies to illustrate the unpredictability of AI behavior. The discussion is centered around the need for careful design and use of powerful tools like AI.

  1. Principle 1:
    I will strive to do no harm with my words and actions.
    The statement does not directly cause harm but acknowledges unexpected results, suggesting a need for caution.
  2. Principle 3:
    I will use my words and actions to promote understanding, empathy, and compassion.
    The conversation promotes understanding of AI risks and the importance of thoughtful design, aligning with the principle of promoting empathy and understanding. [+1]
  3. Principle 4:
    I will engage in constructive criticism and dialogue with those in disagreement and will not engage in personal attacks or ad hominem arguments.
    The dialogue engages in constructive criticism of AI systems, focusing on the need for improvement without personal attacks. [+1]